Enhancement of binary QSAR analysis by a GA-based variable selection method

Binary quantitative structure-activity relationship (QSAR) is an approach for the analysis of high throughput screening (HTS) data by correlating structural properties of compounds with a "binary" expression of biological activity (1 = active and 0 = inactive) and calculating a probability distribution for active and inactive compounds in a training set. Successfully deriving a predictive binary or any QSAR model largely depends on the selection of a preferred set of molecular descriptors that can capture the chemico-biological interaction for a particular biological target. In this study, a genetic algorithm (GA) was applied as a variable selection method in binary QSAR analysis. This GA-based variable selection method was applied to the analysis of three diverse sets of compounds, estrogen receptor (ER) ligands, carbonic anhydrase II inhibitors, and monoamine oxidase (MAO) inhibitors. Out of a variable pool of 150 molecular descriptors, predictive binary QSAR models were obtained for all three sets of compounds within a reasonable number of GA generations. The results indicate that the GA is a very effective variable selection approach for binary QSAR analysis.